a16z-infra / ai-town Goto Github PK

Following the steps on the README and getting pinecone, clerk, and convex all setup for the first time. Characters seem to move around, but things seem to be partially broken. I'm expecting to be able to see conversations happening between the characters at least, but I only see their intro text. There was a message about "sign in to view messages" and after I do that, the "sign in" message disappears, but I don't see any conversations still.

First thing I noticed was an authentication error in the JS, but I DO see my google account photo in the upper right hand corner, so not clear if that's actually an issue. I turned off all shields in brave and repeated the login process, but still get the same thing.

There was an issue with free pinecone I guess (noticed the closed PR related to the error). I un-commented the lines from convex/lib/pinecone.ts with the diff linked in the ticket, which resolved the immediate errors when running npm run dev it seemed, but I'm still getting errors here and there of a similar (cryptic) sort.

I also notice that pinecone has zero vectors, so I'm guessing this is related.

Looking at the embeddings in convex, they were added earlier it seems, but not recently..

Same thing for memories..

Any ideas?

Having the ability add new models/agents using the agent protocol would make it very easy to integrate new agents without having to modify the original codebase.

• https://agentprotocol.ai/
• https://github.com/AI-Engineers-Foundation/agent-protocol

Now that we've sped up the loop with Pinecone some convos are happening too fast to follow, e,g., one paragraph per second. This needs a more natural delay.

Allow parametrising a AI model container and other database endpoint etc

When using my OpenAI API Key, a rate limit error will be reported because the card is not bound. Use the Azure OpenAI API Key to report this error:

failure
Uncaught Error: Embedding failed with code 401: {
    "error": {
        "message": "Incorrect API key provided: 913a2b0d********************0993. You can find your API key at https://platform.openai.com/account/api-keys.",
        "type": "invalid_request_error",
        "param": null,
        "code": "invalid_api_key"
    }
}

Cool project!

It would be neat if this could be run locally / offline without the need for cloud services or subscriptions.

Tools such as LocalAI, Text Generation Web UI, LM Studio provide OpenAI compatible API and functionality - from my experience it's as simple setting the API endpoint.

However with Pinecone - I'm not sure how standard it is, I've seen other projects using Chroma and Weaviate but I'm not sure if they're pluggable.

there's some slightly suboptimal formatting on mobile, but mostly just the wrapping with convex logo at the bottom that i messed up when i changed it :)

Currently map doesn't get reloaded on reset, so if map file changes you have to delete map by hand in the Dashboard. We should fix this :)

Don't see any server side errors. Is this a long OpenAI delay? They eventually unblock. We should make sure a timeout is being enforced.

Installed and got everything running just fine but my agents don't seem to be talking or doing anything. Most are just standing with a chat bubble and Bob is the only one that has any log of doing something (he started a conversation). Been running for about 10 minutes now. Am I missing something?

is ai town launguge model must be using openai's GPT api? or its option ?

i can't tell if my ppl are talking to each other. I have the AI-town running on the local machine but can't read any past convos.

I tried using npx convex run init:reset but then I get this

I am sure I'm doing something very silly wrong -- any idea what that might be?

Two places we can use Convex: store chat history, doing vector search on chat history / background

Q:
Welcome to developing with Convex, let's get you logged in.
? Device name: learn-ai-town
Unexpected Error: RPError: outgoing request timed out after 10000ms
ERROR: "dev:init" exited with 1.

Followed readme, but I'm getting an error when the dev instance starts up and conversations don't seem to progress.

This is the error pop-up

Unhandled Runtime Error
Error: Not Found
Call Stack
s

https://xxxx.clerk.accounts.dev/npm/@clerk/[email protected]/dist/clerk.browser.js (2:134751)

https://xxxx.clerk.accounts.dev/npm/@clerk/[email protected]/dist/clerk.browser.js (2:77226)

Hello,

are there any plans for expanding the world map? E.g. like in the original paper, where characters had their houses, workplaces etc.

If that's in scope, I think it'd be great to have a brief guide into how to develop the map in this project, e.g. creating pixelart, adding it to the game and mapping coordinates to actual places in characters' awareness.

The current version on Main uses the Pinecone Namespaces feature which fails with a very inscrutable error message:

Error: Not implemented: get body for Response: Try using .text(), .json(), blob(), or .arrayBuffer() instead. Consider calling an action defined in Node.js instead 

This happens when running under a Convex V8 action. When run in Node we get the real error message:

Error: PineconeClient: Error calling upsert: PineconeError: The requested feature 'Namespaces' is not supported by the current index type 'Starter'

Technically we don't need the namespaces but we should be careful about folks running Convex dev and prod deployments with the same Pinecone index. We might need to add the deployment name to the index if it doesn't work out of the box.

see this thread for context https://discord.com/channels/1122748573000409160/1142862037970604123/1143052441911312395

Will it consume a lot of keys if the chat is running all the time?

Today playerRelations object in agent.ts contains the wrong data: if the currently speaking player's name is Lucky, playerRelations is referencing Lucky again, making the speaking player thinks they are talking to themselves.

We should fix this so playerRelations contains relationship data between the speaking player and a list of audience they are speaking to.

• explore generating characters using midjourney & add instructions

When the agents respond without capitalizing their own name, which it appears they do occasionally, it gets dropped into the UI in one big text box:

Hey Pete! It's been a while since we last saw each other. How have you been? pete: Hey Stella! I've been great, thank you for asking. How about you? stella: I've been good, just keeping busy. So, I couldn't help but notice how calm and peaceful you always seem. Have you found something that brings you that inner peace? pete: Absolutely! I've been focusing a lot on my faith and spirituality lately. It has made a huge difference in my life. stella: That's wonderful to hear, Pete. I've always admired your dedication and openness to exploring deeper spiritual connections. Could you tell me a bit more about your experiences and how it has positively impacted you? pete: Of course! I feel like my faith has given me a sense of purpose and guidance. It has helped me through tough times and provided me with a community of supportive individuals. Praying and seeking God's guidance has brought me so much peace and joy. stella: That's incredible, Pete. It sounds like you've found something truly special. I genuinely admire your connection to God and how it manifests in your life. I think spirituality can bring so much positivity and light into one's life. pete: Thank you, Stella. It means a lot to hear that coming from you. Spirituality truly is transformative, and I'm grateful for the peace it has brought me. stella: You're very welcome, Pete.

We've encountered a runtime issue with our web application which seems to be related to the @clerk/clerk-js library.

Error encountered: Unhandled Runtime Error
Relevant functions: o._fetch and H.create
Error location: https://suited-cicada-78.clerk.accounts.dev/npm/@clerk/clerk-js@4/dist/clerk.browser.js
While the precise root cause is not immediately clear, it appears there might be issues in the data retrieval process within the o._fetch function or during data creation in H.create.

Could you please investigate further? Additionally, it might be helpful to consult the library's documentation or community for potential solutions.

all agent's history like :

Can I check the conversation without login when runing locally?

The ability to add a character card to every character would be amazing for immersion, and when interacting with them their avatar is shown next to their text conversations.

Not really an issue as such, but you should really credit Pixi as being responsible for all the rendering going on.

I built a server LLaMA 2 in local and can not connect to Convex app.

If this ai town is legit did I make mistake in reading the basics? Like following it to them knowledge to actually have ais go-to the html code of ai town??

I dont know how to fix it please help

WeChat 微信: 18846086541
email : [email protected]

memory stream -> retrieve -> reflect -> plan -> act

npm WARN cleanup Failed to remove some directories [
npm WARN cleanup [
npm WARN cleanup 'D:\summer\ai-town\node_modules\@typescript-eslint\eslint-plugin',
npm WARN cleanup [Error: EPERM: operation not permitted, rmdir 'D:\summer\ai-town\node_modules@typescript-eslint\eslint-plugin\dist\rules'] {
npm WARN cleanup errno: -4048,
npm WARN cleanup code: 'EPERM',
npm WARN cleanup syscall: 'rmdir',
npm WARN cleanup path: 'D:\summer\ai-town\node_modules\@typescript-eslint\eslint-plugin\dist\rules'
npm WARN cleanup }
npm WARN cleanup ],
npm WARN cleanup [
npm WARN cleanup 'D:\summer\ai-town\node_modules\@jridgewell',
npm WARN cleanup [Error: EPERM: operation not permitted, rmdir 'D:\summer\ai-town\node_modules@jridgewell\trace-mapping'] {
npm WARN cleanup errno: -4048,
npm WARN cleanup code: 'EPERM',
npm WARN cleanup syscall: 'rmdir',
npm WARN cleanup path: 'D:\summer\ai-town\node_modules\@jridgewell\trace-mapping'
npm WARN cleanup }
npm WARN cleanup ],
npm WARN cleanup [
npm WARN cleanup 'D:\summer\ai-town\node_modules',
npm WARN cleanup [Error: EPERM: operation not permitted, rmdir 'D:\summer\ai-town\node_modules@flydotio'] {
npm WARN cleanup errno: -4048,
npm WARN cleanup code: 'EPERM',
npm WARN cleanup syscall: 'rmdir',
npm WARN cleanup path: 'D:\summer\ai-town\node_modules\@flydotio'
npm WARN cleanup }
npm WARN cleanup ],
npm WARN cleanup [
npm WARN cleanup 'D:\summer\ai-town\node_modules\next',
npm WARN cleanup [Error: EPERM: operation not permitted, rmdir 'D:\summer\ai-town\node_modules\next\dist\esm\server'] {
npm WARN cleanup errno: -4048,
npm WARN cleanup code: 'EPERM',
npm WARN cleanup syscall: 'rmdir',
npm WARN cleanup path: 'D:\summer\ai-town\node_modules\next\dist\esm\server'
npm WARN cleanup }
npm WARN cleanup ]
npm WARN cleanup ]
npm ERR! code 1
npm ERR! path D:\summer\ai-town\node_modules\hnswlib-node

Hi, wondering if I could put down a CFO like this https://earningsage.lepton.run/ into the town?